REVIEWS
Cognitive neuroscience of emotional memory Kevin S. LaBar and Roberto Cabeza
Abstract | Emotional events often attain a privileged status in memory. Cognitive neuroscientists have begun to elucidate the psychological and neural mechanisms underlying emotional retention advantages in the human brain. The amygdala is a brain structure that directly mediates aspects of emotional learning and facilitates memory operations in other regions, including the hippocampus and prefrontal cortex. Emotion–memory interactions occur at various stages of information processing, from the initial encoding and consolidation of memory traces to their long-term retrieval. Recent advances are revealing new insights into the reactivation of latent emotional associations and the recollection of personal episodes from the remote past. Arousal A dimension of emotion that varies from calm to excitement.
Valence A dimension of emotion that varies from unpleasant (negative) to pleasant (positive), with neutral often considered an intermediate value.
Declarative memory (Or explicit memory). Conscious memories for events and facts that depend on the integrity of the MTL.
Non-declarative memory (Or implicit memory). Various non-conscious memories that are independent of MTL function and are expressed as a facilitation in behavioural performance due to previous exposure.
Center for Cognitive Neuroscience, Duke University, Durham, North Carolina 27708, USA. Correspondence to K.S.L. e-mail:
[email protected] doi:10.1038/nrn1825
Emotional memories constitute the core of our personal history. Philosophers and psychologists have long theorized about how emotion enhances or disrupts memory. Francis Bacon called strong emotion one of the six “lesser forms of aids to the memory”1 and, more recently, Daniel Schacter referred to emotional persistence as one of the seven “sins of memory”2. Over the past century, emotional faculties were analysed primarily through the methods of animal behaviourism and social/clinical psychology, while being eschewed by traditional cognitive psychology. Cognitive neuroscientists have now reversed course to investigate how emotional events are learned and remembered in the human brain. These studies are beginning to elucidate the organization of emotional memory networks at the systems level, providing an important translational bridge between animal models and clinical disorders. Emotion theorists often assume that affective space is parsed according to two orthogonal dimensions — arousal and valence. The impact of these dimensions on different forms of memory, including declarative (explicit) and non-declarative (implicit) memory, has been investigated. The current state of knowledge regarding emotional effects on these memory systems in humans is reviewed below, with an emphasis on arousalmediated influences of the amygdala and its interactions with brain structures in the frontal and temporal lobes3–5 (FIG. 1). The review encompasses several methods of cognitive and affective neuroscience, including studies of patients with medial temporal lobe (MTL) damage, neurohormonal manipulations and functional brain imaging. Within declarative memory, we focus mainly
on memory for events, or episodic memory, and, in the case of non-declarative memory, we focus primarily on fear conditioning, as the greatest advances so far have been made in these areas. Most studies have examined emotional influences under conditions of moderately high arousal, but some studies on the effects of valence in the absence of high arousal are mentioned briefly. Although emotion predominantly benefits memory, long-lasting detrimental consequences are sometimes observed, particularly after severe or prolonged stress (BOX 1). Experimental support for classic views of emotional memory derived from research in non-human animals is described, in addition to findings that expand on this foundation to encompass domains of uniquely human aspects of recollection.
Emotional episodic memory Consequences of amygdala damage. As in other domains of cognitive neuroscience, studies of brainlesioned patients provide a core foundation to delineate structure–function relationships — in this case, determining which aspects of emotional memory depend on the integrity of the amygdala. In humans, organic syndromes rarely affect the amygdala selectively. If the brain damage extends to adjacent MTL memory structures bilaterally, the patient is rendered amnesic, which complicates the study of emotional effects. Therefore, key insights have been provided by post-surgical studies of temporal lobectomy patients with unilateral damage to the MTL due to epilepsy, as well as case studies of rare patients with selective bilateral amygdala pathology due to Urbach–Wiethe syndrome.
54 | JANUARY 2006 | VOLUME 7
www.nature.com/reviews/neuro
© 2006 Nature Publishing Group
REVIEWS Dorsal Working memory Meta-memory Memory strategies Prospective memory Ventral Semantic memory Extinction learning Conceptual priming Autobiographical retrieval
Procedural (habit) learning Reward learning PFC
Striatum
Conditioned emotional learning
HPA axis
Amygdala
MTL memory system Declarative memory Memory consolidation Contextual fear memory Complex conditioning
Cerebellum
Hippocampus Entorhinal cortex Perirhinal cortex
Sensory neocortex
Neurohormonal Reflexive conditioning Memory storage memory modulation Motor learning Conceptual priming Perceptual priming
Figure 1 | Potential mechanisms by which the amygdala mediates the influence of emotional arousal on memory. In addition to emotional learning that takes place intrinsically in the amygdala, direct and indirect neural projections target several memory systems in the brain, including those that subserve working memory, declarative memory and various non-declarative forms of memory (for example, procedural learning, priming and reflexive conditioning). Complex conditioning refers to various higher-order conditioning procedures that are hippocampal dependent, including trace conditioning and conditional discrimination learning. The amygdala also triggers the release of stress hormones by way of the hypothalamic–pituitary–adrenal (HPA) axis, which feed back onto memory consolidation and storage sites as well as the amygdala itself to enhance memory over longer time intervals. Solid arrows indicate direct connections, dashed arrows indicate indirect connections. Blue labels indicate connections with subcortical structures. MTL, medial temporal lobe; PFC, prefrontal cortex.
Recollection Episodic retrieval that is accompanied by recovery of specific contextual details about a past event.
Urbach–Wiethe syndrome (Or lipoid proteinosis). A rare, hereditary, congenital disorder characterized by systemic deposits of hyaline material that are prominent in the skin, oral mucosa and pharynx. About 50% of all cases have additional intracranial deposits in the MTL, which occasionally target the amygdala selectively.
Since the seminal findings of Kleinsmith and Kaplan6, behavioural studies in healthy adults have shown that emotional advantages in memory are sometimes augmented over time. For example, retention advantages for emotionally arousing words relative to neutral words are greater when memory is tested after long (1 h to 1 day) than after short (immediate) delay intervals7,8. Such observations provide evidence that emotional arousal benefits memory in part by facilitating consolidation processes, which take time to emerge. Temporal lobectomy patients do not show enhanced arousal-mediated memory consolidation but instead show parallel forgetting rates for arousing and neutral words from immediate to 1-h retention intervals7,9. However, the exact time course of consolidation is subject to considerable debate, and stabilization of memory traces is a protracted process that could last from months to years. Urbach–Wiethe syndrome patients also show impairments in long-term (1 h to 1 month) recall or recognition of emotional words, pictures and stories10–12 (FIG. 2). Although these results have also been interpreted to reflect deficits in emotion-enhanced consolidation processes, we note that short-term memory for these items has not been tested. Emotional arousal has complementary, immediate effects during encoding that are time invariant and are interpreted to reflect attentional influences on memory3. One additional consequence of emotional arousal is the focusing of attention on central gist information at the
expense of peripheral details for complex events such as emotional narratives or social encounters, as exemplified by ‘weapon focus’ in eyewitness testimony research. Attentional focusing ensures that emotionally salient features of complex events are preferentially retained in memory, which confers evolutionary advantages. Patients with amygdala lesions do not focus on central gist information when memory is tested for audiovisual narratives that describe emotionally arousing events13. The gist memory effects are evident when the patients generate intact skin conductance responses (SCRs) and arousal/valence ratings to the stimuli, which implicates impairments in emotion–cognition interactions rather than a more basic problem with emotional evaluation. In contrast to the above findings, certain aspects of emotional memory are preserved following amygdala damage. Patients with amygdala lesions do preferentially remember words that are affectively valent but low in arousal relative to neutral ones, as well as neutral words encoded in emotional sentence contexts relative to neutral contexts9,14. In such cases, it is possible that the patients access other cognitive resources that boost retention, including semantic cohesion15 and organizational strategies, which are probably mediated by direct interactions between the MTL and the prefrontal cortex (PFC). Some emotional benefits in memory are therefore possible without a fully functioning amygdala, especially when damage occurs later in life16, and relate in part to recruitment of cognitive processes by emotional valence. These findings extend the results from research in rodents to suggest that arousal rather than valence is the crucial factor in engaging the amygdala during emotional memory tasks17. Neurohormonal memory modulation. A limitation of most patient studies is that impaired performance observed at a single time point could reflect deficits in any one of four memory stages: encoding, consolidation, storage or retrieval. Psychopharmacological studies in healthy adults can provide somewhat greater specificity regarding which phase of memory is affected by emotion. However, the prolonged time course of neurohormonal actions must be considered, as stress hormones can have different effects on each stage of memory. As shown in rodents by McGaugh and colleagues18,19, adrenal stress hormones modulate performance on various learning and memory tasks. Emotional situations initiate complex interactions between adrenergic and glucocorticoid systems that are coordinated by the hypothalamic–pituitary–adrenal axis at central and peripheral sites of action. Adrenaline release in the periphery stimulates vagal afferents that terminate in the nucleus of the solitary tract, which, in turn, projects to the amygdala and other memory-related forebrain regions. Posttraining infusion of β-adrenergic receptor antagonists into the basolateral amygdala blocks the memory-enhancing effects of adrenaline, and infusion of β-adrenergic receptor agonists facilitates memory consolidation20,21. Within the basolateral amygdala and hippocampus, noradrenaline enhances glutamatergic synaptic plasticity, which is thought to underlie learning and memory functions22,23.
NATURE REVIEWS | NEUROSCIENCE
VOLUME 7 | JANUARY 2006 | 55
© 2006 Nature Publishing Group
REVIEWS Box 1 | Post-traumatic stress disorder Post-traumatic stress disorder (PTSD) emerges after exposure to a traumatic stressor that elicits fear, horror or helplessness and involves bodily injury or threat of injury or death to one’s self or another person. Community-based studies in the United States estimate a lifetime prevalence of trauma exposure at 50%, but only 5% of men and 10% of women will develop PTSD139. Prevalence rates are higher in at-risk populations, such as war veterans. Diagnostic symptoms include persistent re-experiencing of the traumatic event, avoidance of reminders, numbing of responsiveness and heightened arousal. Neurobiological models of PTSD have focused on brain regions and stress hormone systems that are involved in fear, arousal and emotional memory. Cortisol dysregulation and abnormal responses to adrenergic modulators implicate disturbances in the hypothalamic–pituitary–adrenal axis and its interactions with brain regions that control arousal140. Chronic stress in PTSD contributes to smaller hippocampal volume and declarative memory deficits141. Symptom provocation studies show blood flow changes in cortico-limbic circuitry involved in emotional memory, including the amygdala, anterior cingulate and orbitofrontal cortex142,143. Patients with PTSD have exaggerated startle responses to loud sounds144 and show greater contextual and cued fear conditioning145,146. Given that administration of the β-adrenergic receptor antagonist propranolol selectively reduces memory consolidation for emotionally arousing material (FIG. 2), beta-blockers are currently being evaluated as potential agents for secondary prevention of PTSD147. However, the ethics of this approach, as well as its empirical and theoretical basis, is still the subject of some debate.
Working memory A form of memory in which stimulus representations are actively maintained and/or manipulated in conscious awareness over a short period of time.
Memory consolidation for both appetitively and aversively motivated learning tasks is blocked by adrenocortical suppression, and is enhanced by infusions of glucocorticoid receptor agonists into the basolateral amygdala and hippocampus24–26. Lesions of the basolateral amygdala modulate the effectiveness of glucocorticoid manipulations in the hippocampus, which implicates a functional coupling between these regions for arousal-enhanced memory consolidation25. Because the behavioural impact of stress and glucocorticoids is modulated by α- and β-adrenergic receptor activation in the basolateral amygdala, the mnemonic effects of different stress hormone systems are co-dependent27,28. Collectively, the findings from research in rodents have been interpreted as evidence to support the memory-modulation hypothesis, which states that greater long-term memory for emotional than neutral events reflects the neuromodulatory influence of the amygdala on consolidation processes in MTL memory regions through engagement of stress hormones18. Analogous neurohormonal mechanisms also contribute to amygdala influences over other memoryprocessing regions of the brain, although the behavioural consequences are not always advantageous. For example, acute corticosterone administration impairs performance on tests of working memory, which also depend on adrenaline and on interactions between the amygdala and PFC29. Investigations with humans have begun to examine neurohormonal influences across emotional and nonemotional memory tasks. In patients with epilepsy, recognition memory for prose passages is enhanced following moderate-intensity stimulation of the vagus nerve, which provides a route for peripheral hormones to feed back onto central learning sites30. Pharmacological manipulations in humans have implicated both adrenergic and corticosteroid influences on memory, although with less anatomical specificity than in animal studies. Administration of βadrenergic receptor antagonists (for example, propranolol)
before encoding reduces the long-term retention advantage typically seen for emotionally arousing stimuli relative to neutral stimuli31,32 (FIG. 2). Conversely, administration of β-adrenergic receptor agonists (for example, yohimbine) promotes emotional memory33,34. However, one study found that 40 mg of propranolol does not affect emotional memory when administered after encoding35, and another study found impairing effects of a higher dose (80 mg) of propranolol across both short- and long-term retention intervals36. During short retention intervals, propranolol also induces a retrograde amnesia for neutral words that precede emotional words in a list, an effect that is larger in females37. Additional research in humans is warranted to dissociate adrenergic modulation of attentional effects during encoding (that affect both short- and long-term memory) from consolidation effects, which have been shown consistently in rodents. Comparison of adrenergic receptor antagonists that readily cross the blood–brain barrier (for example, propranolol) versus those that do not (for example, nadolol) shows that β-adrenergic effects on emotional memory in humans are mediated by receptors located in the brain32 (but see REF. 38 for an alternative point of view). Neuropsychological and functional neuroimaging studies have converged to identify the amyg dala as a likely mediator of these influences. On tests of memory for emotionally arousing words or stories, damage to the amygdala yields impairments similar to those of beta-blocker administration in healthy controls9,11,12,37,39 (FIG. 2). In addition, functional neuroimaging studies indicate that amygdala activity during the encoding of emotional stimuli is reduced by propranolol40,41, with a concomitant reduction in hippocampal activity during retrieval of the same stimuli40. Stress and glucocorticoids affect both emotional and non-emotional forms of memory in humans. During encoding, acute cortisol administration or stress-induced endogenous cortisol release generally enhances emotional learning and memory42–45, but similar manipulations during retrieval impair recall of earlier memories46,47. The acute impact of cortisol is typically greater for emotionally arousing stimuli than it is for neutral stimuli, although some studies have found similar effects across emotional and neutral material48, or effects in the opposite direction49,50. On tests of working memory, psychosocial stress or highdose cortisol administration typically impair performance, which is consistent with the animal literature51–53. Variations in cortisol influences on memory are attributable to several factors, including biological sex, duration of stress (acute versus chronic), cortisol dose (typically as an inverted U-shaped function) and time of day relative to the circadian flux in endogenous cortisol levels18,19. Variations caused by cortisol dose and time of day relative to circadian flux are related to the relative occupancy of mineralocorticoid or glucocorticoid receptor subtypes, which have different affinities for glucocorticoids and affect memory functions to different extents. At low doses, mineralocorticoid receptor activation is dominant and relates to emotional enhancement of encoding processes, but consolidation benefits are typically not reported. At higher doses, glucocorticoid receptor activation,
56 | JANUARY 2006 | VOLUME 7
www.nature.com/reviews/neuro
© 2006 Nature Publishing Group
REVIEWS 80
Memory score
70 60 50 40 30 20
Controls: placebo Controls: propranolol Patients Neutral
Emotional
Neutral
Story phase
Figure 2 | β-Adrenergic receptor blockade in healthy adults during encoding produces similar deficits to amygdala damage on a test of emotional memory. Participants view a slide show and hear an accompanying narrative. The middle portion of the story describes a car accident, whereas the beginning and end portions of the story are emotionally neutral in content. Healthy adults given a placebo 1 h before the story remember the emotionally arousing portion of the story better than the neutral portions 1 week later. Propranolol administration during encoding abolishes this retention advantage in healthy adults. Two patients with selective amygdala damage (SM and BP, data averaged) also lack the retention advantage for the emotionally arousing portion of the story. Modified, with permission, from REF. 12 © (1997) Cold Spring Harbour Laboratory Press and REF. 31 © (1994) Macmillan Magazines Ltd.
combined with adrenergic influences, contributes to enhanced memory consolidation. In contrast to acute effects, chronic elevations in basal cortisol levels in older high-stress individuals54 or altered stress reactivity in some neuropsychiatric disorders, such as depression55 and post-traumatic stress disorder (BOX 1), can lead to reductions in hippocampal volume and concomitant declarative memory deficits, even for non-emotional material. Cortisol-induced impairments in declarative memory retrieval have been linked to reductions in MTL activity56. However, as mentioned above with respect to working memory, stress hormone systems project to a diffuse set of brain areas (including the PFC, cerebellum, hypothalamus and hippocampus), each of which are subject to modulation by the amygdala for different memory operations with potentially distinct consequences (FIG. 1).
Dm effect An index of brain activity at encoding that distinguishes subsequently remembered from subsequently forgotten items and is assumed to reflect successful encoding processes.
Imaging emotional memory encoding. Brain imaging using positron emission tomography (PET) and functional MRI (fMRI) can also distinguish the impact of emotion at different stages of episodic memory, but with far superior specificity of the underlying neuroanatomy than hormonal manipulations. Moreover, these studies have the potential to reveal functional interactions among distributed brain regions to test the memory-modulation hypothesis and to implicate involvement of additional areas. Most functional imaging studies on this topic have investigated encoding processes57–67. Consistent with the memory-modulation hypothesis, activity
in the amygdala and MTL memory regions during the encoding of emotional stimuli is correlated with individual differences in later memory for these stimuli. PET studies initially established that the amount of amygdala activation during encoding positively correlates with delayed recall accuracy for aversive but not neutral film clips57, as well as delayed recognition accuracy for emotionally arousing pictures that are both positive and negative in valence59. In addition, there is a sex difference in the hemispheric distribution of encoding-related amygdala activity, with men showing right-lateralized effects and women showing left-lateralized effects61. The sexually dimorphic lateralization pattern is more prominent when considering the relationship between amygdala activity and memory, and is found less often as an effect of emotion on perceptual processing. Reasons for the sex difference in emotional memory remain unclear and constitute an active area of current research. Event-related fMRI experiments have replicated the correlation between amygdala activity at encoding and delayed retention accuracy for emotional pictures, as well as the dependency of this relationship on self-reported arousal levels and lateralization by sex60,62,63. Confirmation by this technique is important because the temporal resolution of PET and fMRI when using blocked designs does not permit researchers to distinguish transient emotional effects from sustained mood influences, and individual items cannot be analysed according to emotion ratings or memory performance (for example, retrieval success or failure) obtained from each participant. Furthermore, event-related designs allow the use of the subsequent memory paradigm, which can distinguish activity associated with stimulus processing and task demands to reveal neural signatures that specifically reflect successful encoding operations. This paradigm isolates the Dm effect by contrasting study-phase activity for items that are remembered versus those that are forgotten in a subsequent memory test68. The enhancing influence of emotion on successful encoding activity can then be investigated by comparing the Dm effect for emotional versus neutral stimuli. For example, an event-related potential (ERP) study showed that the Dm effect for emotional stimuli occurred faster (400–600 ms) than the Dm effect for neutral stimuli (600–800 ms), suggesting that emotional stimuli have privileged access to processing resources69. Consistent with the memory-modulation hypothesis, fMRI studies have shown that emotion enhances the Dm effect in both the amygdala and MTL memory regions60,62,64,65,70 (FIG. 3). In addition, the Dm effect for emotional and neutral items differs in localization within the MTL, with the former being mediated in anterior parahippocampal regions and the latter in posterior parahippocampal regions64. This functional localization is consistent with anatomical evidence for greater reciprocity between the amygdala and anterior sectors of the parahippocampal gyrus71,72. Although MTL structures have been emphasized in support of the memory-modulation hypothesis, we note that the PFC also contributes to emotional Dm effects, with regionally specific modulation by both arousal and valence65,67,73 (FIG. 3).
NATURE REVIEWS | NEUROSCIENCE
VOLUME 7 | JANUARY 2006 | 57
© 2006 Nature Publishing Group
Left hippocampus
Left inferior PFC
Peak percentage signal change
Left amygdala
0.4 0.35 0.3
0.4
Peak percentage signal change
a
Peak percentage signal change
REVIEWS * Remember Forget
0.25 0.2 0.15 0.1 0.05 0
0.35
Arousal *
Valence only
Neutral
* *
0.3 0.25 0.2 0.15 0.1 0.05 0
Arousal
0.35
Valence only *
0.3
Neutral
*
0.25 0.2 0.15 0.1 0.05 0
Arousal
Valence only
Neutral
b Arousal Dm
BA 47
BA 6/9
Arousal Dm
Neutral Dm
Figure 3 | Two routes to emotional remembering: arousal- and valence-mediated subsequent memory effects. a | Functional MRI (fMRI) activation is monitored while healthy adults encode high-arousing negative words, low-arousing negative words (valence only) and neutral words. For each participant, the data from each trial are sorted off-line according to whether the word is subsequently remembered or forgotten to generate an index of successful encoding (called the Dm effect or subsequent memory effect). Relative to neutral words, high-arousing negative words generate greater Dm effects in the hippocampus and amygdala, whereas low-arousing negative words generate greater Dm effects in the hippocampus and a posterior region of inferolateral prefrontal cortex (PFC). Asterisks indicate significant differences (p